Razorpay, an omnichannel payments platform for businesses, has launched Vulcan, an AI foundation model for payments developed with NVIDIA and AWS.
Vulcan is a transformer-based AI foundation model built specifically for payments processing. The model was developed using NVIDIA's accelerated computing technology and AWS's cloud infrastructure, including Amazon SageMaker, and is intended to improve the reliability, security, and predictability of digital transactions across India's payments ecosystem.
According to Razorpay, an internal study covering 1.5 million shoppers and more than 51.000 businesses found similar patterns of payment friction, including failed transactions and drop-offs, occurring consistently across both metropolitan and small-town markets. This finding informed the company's decision to build a single shared model rather than maintain separate systems for routing, fraud detection, risk assessment, and checkout, which the company said do not currently share data with one another despite relying on similar signals.
Technology and training
Vulcan is built on transformer architecture, the same category of technology underpinning large language models, like LLMs, though Razorpay has stated the model is adapted to interpret patterns in payments data rather than text.
The company said the model has been trained on approximately three trillion data points drawn from four billion payments, learning from around 3.000 signals per transaction. Razorpay described the model as proprietary, with both the architecture and training data developed internally.
Early performance data
Ahead of the full launch, early components of the model had already been applied to live transactions on Razorpay's network, with customers including Blinkit, Bachatt, and redBus among those using the capabilities.
Razorpay reported an improvement of 8-10% in payment success rates, an eightfold increase in detected international card fraud, and a fivefold increase in identified fraudulent or disputed transactions without a corresponding rise in alert volumes. The company also said that 40% more shoppers have shown their preferred UPI app on Razorpay Magic Checkout product, which it linked to between 100.000 and 200.000 additional monthly purchases.
Market context
Razorpay positioned the launch within the broader context of India's payments landscape, where a single transaction may be processed through UPI, cards, net banking, wallets, or cash on delivery across numerous banks and gateways.
In addition, Harshil Mathur, CEO & Founder of Razorpay, adds that India’s need for a digital payments industry is real, but going digital is still a challenge for a large part of the country, as putting trust in a phone screen over the traditional cash in hand, is a decision many individuals are not comfortable making yet.
The company said India's ecommerce market is projected to reach USD 350 billion by 2030, and it views Vulcan as part of the infrastructure required to support that growth.
